Maybe just zero-extend the 32-bit value? (or truncate for 16-bit?) I
don't know what generates better machinecode.
On Fri, Jan 12, 2018 at 2:34 AM, Timothy Arceri <tarc...@itsqueeze.com> wrote:
> ---
>
> Open to suggestions for better ways to do this.
>
> src/amd/common/ac_nir_to_llvm.c | 13 ++++++++++---
> 1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
> index b0a8a94a2d..f013b4b387 100644
> --- a/src/amd/common/ac_nir_to_llvm.c
> +++ b/src/amd/common/ac_nir_to_llvm.c
> @@ -1422,9 +1422,16 @@ static LLVMValueRef emit_f2b(struct ac_llvm_context
> *ctx,
> }
>
> static LLVMValueRef emit_b2i(struct ac_llvm_context *ctx,
> - LLVMValueRef src0)
> + LLVMValueRef src0,
> + unsigned bitsize)
> {
> - return LLVMBuildAnd(ctx->builder, src0, ctx->i32_1, "");
> + LLVMValueRef result = LLVMBuildAnd(ctx->builder, src0, ctx->i32_1,
> "");
> +
> + if (bitsize == 32)
> + return result;
> +
> + LLVMValueRef icond = LLVMBuildICmp(ctx->builder, LLVMIntEQ, result,
> ctx->i32_1, "");
> + return LLVMBuildSelect(ctx->builder, icond, ctx->i64_1, ctx->i64_0,
> "");
> }
>
> static LLVMValueRef emit_i2b(struct ac_llvm_context *ctx,
> @@ -1979,7 +1986,7 @@ static void visit_alu(struct ac_nir_context *ctx, const
> nir_alu_instr *instr)
> result = emit_f2b(&ctx->ac, src[0]);
> break;
> case nir_op_b2i:
> - result = emit_b2i(&ctx->ac, src[0]);
> + result = emit_b2i(&ctx->ac, src[0],
> instr->dest.dest.ssa.bit_size);
> break;
> case nir_op_i2b:
> src[0] = ac_to_integer(&ctx->ac, src[0]);
> --
> 2.14.3
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
